Best fit model predictive performance of RRV notifications and outbreaks in local government areas (LGA) in Victoria (VIC), and Western Australia (WA) by LGA climate.

The total number of RRV notifications (Cases), the best model used for predicting RRV notifications, adjusted R-squared coefficient (R2), the best model used for predicting outbreaks, sensitivity (Sn), specificity (Sp), and Matthews correlation coefficient (MCC). ARIMA = auto-regressive integrated moving average model; GAM = generalised additive model; BR = generalised boosted regression; NB = negative binomial regression; and Hurdle = hurdle regression. Ninety five percent confidence intervals (95% CI) are given of the distribution of each predictive performance measure from Jackknife pseudo-random sampling using the respective best fit model. Models with a “*” following the model type used the Factorial Approach. See Table 2 for a comparison of how close modelling methods were to one another for predicting RRV notifications and outbreaks.